Please use this identifier to cite or link to this item:
https://hdl.handle.net/11681/10825| Title: | Migration of WAM to scalable computing environments |
| Authors: | Coastal and Hydraulics Laboratory (U.S.) West, John E. Jensen, Robert E. Turcotte, Louis H. |
| Keywords: | Domain decomposition Operational forecast High-performance computing Pipeline parallelism Spectral wave forecast Message passing MPI Ocean waves Coastal waves Forecasting |
| Publisher: | Information Technology Laboratory (U.S.) Engineer Research and Development Center (U.S.) |
| Series/Report no.: | Technical report (U.S. Army Engineer Waterways Experiment Station) ; ITL-97-6. |
| Description: | Technical Report Abstract: Recent reviews in Department of Defense (DoD) strategic interests place additional emphasis on near-shore, or littoral, operations. This has created a significant Army interest in coastal wave predictions for sustainment of operations and has significantly increased the Navy's needs for improved coastal wave predictions. Both services use a European-developed code called the Wave Model, or WAM, for oceanic wave prediction. Although initially developed for deep ocean, global forecasts, research indicates that extension of WAM to handle near-shore forecasts may be very successful, if adequate computational power is available. This report documents results of work currently underway as part of the DoD's scalable software initiative to improve the performance of WAM by a factor of two to four times current operational performance. Techniques to achieve this level of performance on the emerging breed of scalable, commodity-processor based parallel computing systems in a way which is both efficient and portable are detailed and placed in the context of an overall design philosophy. Development in two phases is discussed. Phase I focuses on the parallelization of the simulation job stream so that each of the successively finer grid nests used in a forecast can be computed simultaneously. Phase II development extends this parallelism by performing computations for each grid on multiple processors. Preliminary results are presented which demonstrate excellent performance improvements in initial evaluations, and specific recommendations for further enhancements are provided. |
| Rights: | Approved for public release; distribution is unlimited. |
| URI: | http://hdl.handle.net/11681/10825 |
| Appears in Collections: | Technical Report |
